Crate brainwires 

Brainwires

The Brainwires Agent Framework — build any AI application in Rust.

Re-exports all framework sub-crates via feature flags for convenient access.

Quick Start

[dependencies]
brainwires = { version = "0.10", features = ["full"] }

use brainwires::prelude::*;

Modules

a2a

A2A (Agent-to-Agent) protocol support.

agent_network

Agent network — IPC, remote bridge, mesh networking, routing, discovery.

agents

Agent runtime, communication hub, task management, and validation.

audio

Audio — capture, playback, speech-to-text, text-to-speech.

camera

Camera and webcam frame capture.

chat

Chat provider implementations (Provider trait wrappers over API clients).

core

Core types and traits — available via brainwires::core::* or brainwires::prelude::*.

datasets

Training data pipelines — JSONL, format conversion, tokenization, dedup.

dream

Offline memory consolidation — summarization, fact extraction, hot/warm/cold tier transitions.

eval

Evaluation framework — Monte Carlo runner, Wilson CI, adversarial tests.

hardware

Hardware I/O — audio, GPIO, Bluetooth, camera, USB, voice assistant.

inference

LLM-driven workhorses — chat agent, task agent, planner / judge / validator helpers, cycle orchestrator, summarization, system prompts.

interpreters

Sandboxed code interpreters — Rhai, Lua, JavaScript (Boa), Python (RustPython).

knowledge

Central knowledge — BKS, PKS, entity graphs, thought processing.

lan

LAN inspection tooling — moved from brainwires-hardware::network into brainwires-network::lan.

mcp

MCP client — connect to external MCP servers and use their tools.

mcp_server_framework

MCP server framework — build MCP-compliant tool servers with middleware.

mcp_server_support

Re-exports for building MCP servers (rmcp, schemars, CancellationToken).

mdap

MDAP — Multi-Dimensional Adaptive Planning with MAKER voting.

memory

Tiered hot/warm/cold agent memory — message/summary/fact stores (schema, from brainwires-stores) plus the TieredMemory orchestration (from brainwires-memory, when the tiered feature is on).

mesh

Distributed mesh networking — topology, discovery, federation, routing.

permissions

Permissions — capability profiles, policy engine, audit logging.

prelude

Convenience prelude — import everything commonly needed.

prompting

Adaptive prompting — technique library, clustering, temperature optimization.

providers

AI provider implementations — OpenAI, Anthropic, Google, Ollama, and more.

rag

RAG — codebase indexing, semantic search, and retrieval-augmented generation.

reasoning

Reasoning — planners, validators, routers, strategies, output parsers.

seal

SEAL — Self-Evolving Adaptive Learning for coreference and knowledge.

skills

Skills — SKILL.md parsing, skill registry, and execution.

storage

Persistent storage primitives — StorageBackend trait, embeddings, BM25.

system

Generic OS-level primitives — filesystem event reactor, service management.

telemetry

Telemetry — analytics events, billing hooks, SQLite persistence, and cost/usage queries.

tools

Model tools — both halves at one path. Runtime types (ToolExecutor, ToolRegistry, error taxonomy, validation, transactions, smart router, plus optional orchestrator / OAuth / OpenAPI / sandbox / sessions / RAG-tool modules) and the concrete builtins (BashTool, FileOpsTool, GitTool, WebTool, SearchTool, BuiltinToolExecutor, registry_with_builtins, plus optional code_exec / interpreters / semantic_search / browser / email / calendar / system).

training

Model training — cloud fine-tuning, local Burn-based LoRA/QLoRA/DoRA.

usb

Raw USB device access and transfers.

vad

Voice Activity Detection.

voice_assistant

Voice assistant pipeline.

wake_word

Wake word detection.